Skip to main content

The drivers and functions of rock juggling in otters

Cite this dataset

Allison, Mari-Lisa; Reed, Rebecca; Michels, Emile; Boogert, Neeltje (2020). The drivers and functions of rock juggling in otters [Dataset]. Dryad.


Object play refers to the seemingly non-functional manipulation of inanimate items when in a relaxed state. In juveniles, object play may help develop skills to aid survival. However, why adults show object play remains poorly understood. We studied potential drivers and functions of the well-known object play behaviour of rock juggling in Asian small-clawed (Aonyx cinereus) and smooth-coated (Lutrogale perspicillata) otters. These are closely related species, but Asian small-clawed otters perform extractive foraging movements to exploit crabs and shellfish while smooth-coated otters forage on fish. We thus predicted that frequent rock jugglers might be better at solving extractive foraging puzzles in the first species, but not the latter. We also assessed whether species, age, sex and hunger correlated with rock-juggling frequency. We found that juvenile and senior otters juggled more than adults. However, rock-juggling frequency did not differ between species or sexes. Otters juggled more when “hungry”, but frequent jugglers did not solve food puzzles faster. Our results suggest that rock juggling may be a misdirected behaviour when hungry and may facilitate juveniles’ motor development, but it appears unrelated to foraging skills. We suggest future studies to reveal the ontogeny, evolution and welfare implications of this object play behaviour.


Study sites and test subjects

We collected data at three sites in the United Kingdom: New Forest Wildlife Park, Newquay Zoo and Tamar Otter and Wildlife Centre. At New Forest Wildlife Park, four groups of Asian small-clawed otters (“ASC”; N = 4, 4, 2, 1) and two groups of smooth-coated otters (“SCO”: N = 4, 2) were studied. Newquay Zoo held one group of Asian small-clawed otters (N = 12). Tamar Otter and Wildlife Centre had three groups of Asian small-clawed otters (N = 15, 3, 3). Across the three sites, we studied a total of 23 males (ASC: N = 21; SCO: N = 2) and 27 females (ASC: N = 23; SCO: N = 4). Ages ranged from 6 months to 19 years for Asian small-clawed otters and 3 months to 5 years for smooth-coated otters (for group compositions, see supplementary materials table S1).

To address the research questions on individual and species differences in rock juggling frequency, we first collected observational data on each individual to determine when and how often rock juggling was performed. To address whether rock juggling facilitates extractive foraging behaviour, we then presented each otter group with novel extractive food puzzles to solve.

Rock-juggling observations

Prior to data collection, we carried out preliminary observations to compile an ethogram (supplementary materials table S2).

Data collection dates and times differed between study sites due to the location and availability of the establishment and feeding times. At New Forest Wildlife park, we collected data between August 5th and September 4th 2018, between 1000 – 1400 hours. At Newquay Zoo, we collected data between October 17th 2018 and February 14th 2019, between 1030 – 1230 hours and 1430 – 1630 hours. At Tamar Otter and Wildlife Centre, we collected data between December 10th 2018 and January 15th 2019, between 0900 – 1630 hours. We identified otters using differences in body size, shape, fur colour and nose patterns. We conducted observational scans every 3 minutes over a one-hour period. This was repeated to give a total of 12 hours of observation per otter. To assess the influence of hunger levels, we constructed a schedule to observe each otter prior, during and post the establishment’s regular feeding times (supplementary materials table S3). Otters were deemed to be satiated for two hours post feed, as food is reported to take two hours to pass from mouth to spraint in otters (Heap et al. 2008).

Den Camera Trap

To ensure we did not bias our rock juggling observations by only watching otter behaviour outdoors, a motion-triggered camera trap was installed in the indoor enclosure at Newquay Zoo between July 25th and 28th 2018. Each motion-triggered recording was 15 seconds in length and contained a date and time stamp. The “hunger level” of the otters could thus be inferred by time since last feed. Due to the low resolution of the recordings, we could not ID the otters. Body size was used to differentiate adults from pups; otters ≤1 year old were considered pups. We scored the number of adult otters and pups present, and whether individuals were showing rock juggling.

Novel Extractive Food Puzzles

We designed three novel extractive food puzzles to quantify individual variation in extractive foraging performance. These puzzle types were designed to be novel to the otters, to control for prior experiences that could influence results and included: (a) 50ml white opaque screw-top medicine bottles (radius=~2cm, height=~7cm including lid), (b) tennis balls with an 8cm diameter cross cut in, with a 1.5 x 1.5cm square hole in the middle of the cross and (c) pairs of bright green Duplo® bricks (each brick measuring: 3.1cm x 6.4cm x 2.4cm) stacked on top of each other (Figure 1).

We filled each puzzle with 10g of 5% fat lean minced meat. We placed mince inside the bottles and loosely secured the lid to test general dexterity and digit strength. The same experimenter closed all bottles to ensure consistency. We placed mince inside the hole of the tennis balls to replicate foraging in nooks and crannies. We stuffed mince between the upper and lower Duplo® brick to mimic the extraction of mussels or clams. We also placed monkey nut shells between the bricks to prevent the bricks from being closed too tightly by the otters. This produced a consistent gap of ~1mm between the bricks.

The puzzle types were presented to each group in a random order. We provided twice as many individual puzzles as there were otters in each study group (e.g. groups of four otters were presented with eight puzzles) to prevent individuals from monopolising the puzzles and increase the opportunity for all otters to solve the puzzles. Puzzles were dropped in at three separate points at each otter enclosure to further facilitate all individuals accessing the puzzles and to aid in individual identification for data collection. At the start of every session, we filled three puzzles in sight of the otters to demonstrate that there was food inside to motivate the otters to extract it.

Puzzles were presented ca. one hour before a feed to ensure that otters were both active and food motivated. Puzzle trials lasted until all puzzles were “solved”, i.e. the food reward was extracted, or until 20 minutes had elapsed. Each puzzle session was recorded with two Panasonic video cameras positioned around the circumference of the enclosure. An additional camera (Fujifilm Finepix HS20 EXR) was used freehand to facilitate the identification of each otter. From each puzzle trial video, we scored each otter’s: (i) latency to first interact with a puzzle, (ii) time spent interacting with a puzzle until the otter solved it and (iii) total time from puzzles being dropped into the enclosure until the otter solved a puzzle. We noted if an otter did not solve the puzzle, in which case the “total time” was the same as the trial duration (1200 seconds).

Statistical Analyses

Rock Juggling Observational Data

To address whether otter species, age or sex were associated with differences in rock juggling frequency, we first summed all observations of rock juggling to provide a total count of rock juggling observations for each otter. The total number of behaviour scans per otter (regardless of behaviour performed) was also noted, as we could not observe each individual for a total of 252 times (i.e. every 3 min for 12 hours, started at 0 min). Missing observations were due to unforeseen circumstances, such as the transfer of an individual to another zoo. The otter with the lowest observation count of <230 behaviour scans was removed from the sample. An extreme outlier with an unusually high propensity to rock juggle (i.e. a total count of 59 while across all others, average±SD = 12.77±14.95) was also removed from analyses. Asian small-clawed otters at the New Forest Wildlife Park were considerably older (all >11 years old) than Asian small-clawed otters at both Newquay Zoo and Tamar Otter and Wildlife Park (all ≤11 years old). To account for this confounding factor in analyses, we categorised otters as either “young” (≤11 years) or “old” (>11 years). This age threshold was informed by the 11 year average life expectancy of captive Asian small-clawed otters (Crandall, 1964) and the reproductive status of the otters to reflect the senescent individuals in the sample.

To test how species, age and sex were associated with rock juggling frequency, we used a negative binomial generalized linear model as clumping in the count data resulted in an over-dispersed Poisson model. An ‘offset’ for total observation count was included to inform the model of variance in observation effort. Rock juggling count was the response variable and species, age, age category, sex and site were the fixed effects. Site was included in the model to account for variation in otter behaviour between the zoos. The model also included interactions between age and age category, and age and site. Stepwise backwards elimination of least significant terms was conducted to produce a minimal adequate model (MAM). For each non-significant predictor, the z- or t- and p-values are reported from the model summary before it was removed from the model, while for significant predictors we also report the slope estimates ± SE from the MAM summary. For each significant predictor, a likelihood ratio test using command ‘anova’ was run to compare the MAM to a model where that fixed effect was removed. These ANOVAs provided the likelihood ratio test/Chisquare and p-values reported and indicate whether there is a statistical difference in the amount of deviance explained by the two models including vs. excluding the predictors of interest.

To test whether hunger levels (as inferred by time since last feed) were associated with differences in juggling frequency, we summed the number of rock juggling observations when otters were assumed to be hungry vs. full, as well as the total number of behaviour scans when otters were hungry/full. A negative binomial generalized linear mixed-effects model was run that offset total observation count, had rock juggling as the response variable, hunger level as the fixed effect and individual ID as a random effect. Model selection was conducted as described above.

Den Camera Trap Data

The camera trap generated 556 usable 15-sec video clips. To test whether hunger levels were associated with rock juggling frequency in Newquay Zoo otters whilst they spent time inside, we used known feeding times and the time of the recording to infer whether otters were relatively hungry or full by calculating time since last feed. As individuals could not be identified, a Pearson’s chi-squared test was run on a contingency table of rock juggling occurrences vs. absences when hungry vs. full. Observed values were then compared to expected values to assess the direction of the effect.

To test whether pup presence influenced rock juggling frequency in adults (e.g. because adults might demonstrate rock juggling to their pups), a Pearson’s chi-squared test was run on a contingency table of rock juggling occurrences vs. absences when pups were present vs. absent. Observed values were then compared to expected values to assess the direction of the effect.

To control for confounding factors, the influence of time since last feed (i.e. “hunger”) on pup presence in the den was assessed. A Pearson’s chi-squared test was run on a contingency table of pup presence vs. absence when otters were relatively hungry vs. full. Observed values were compared to expected values to assess the direction of the effect.

Novel Extractive Food Puzzle Data

To address whether rock juggling frequency was associated with extractive foraging performance on the food puzzles, we first removed an outlier from the dataset. This outlier represented an individual that was present in the enclosure but did not interact with any of the tennis ball puzzles for the duration of the trial and thus had a latency of 1200 seconds, while all other otters had latencies of 90 seconds or less. In addition, for some video trials individuals could not be identified or moved out of camera shot and so did not have solving success, latencies or interaction times accurately recorded. These NAs were removed from the dataset before analyses. As rock juggling was used as a fixed effect in analyses to predict puzzle solving performance, an offset could not be used to control for variance in observer effort. Instead, a juggle rate was calculated by dividing the total count of rock juggling by the total number of observation points for each otter. Personal observation suggested that juggle rate differed between species, and this species difference was found to be significant. We therefore included this juggle rate-species interaction in the statistical models of puzzle interaction data. However, we omitted it from the puzzle solve analyses as it caused major model convergence issues.

To test whether latency to first interaction with each food puzzle type was shorter for younger, and possibly more explorative, otters, and/or for those who rock juggled more frequently, we used a linear mixed-effects model. The response variable was each otter’s latency to first interact with each food puzzle, fixed effects were juggle rate, species and their interaction, age, sex, puzzle type and site, and random effects were individual ID nested within group ID. As plotting the data suggested that the species showed large differences in their latency to start interacting with the different puzzle types, we explored these patterns further by running the same model again, but on the data for each species separately.

Solving success was noted as a binary category with “1” representing success while “0” represented an otter’s failure to solve a puzzle (i.e. extract the food reward) during the trial. To test whether otters that rock juggled at a higher frequency were more successful at solving puzzles, we initially used a generalized linear mixed-effects models with binomial error structure. Solving success was the response variable, fixed effects were juggle rate, species and their interaction, latency to first interaction with the task, age, sex, puzzle type and site, and individual ID nested within group ID were random effects. However, this model failed to converge until we simplified it to include only the main predictors of interest: species, juggle rate and puzzle type.

To test whether otters that rock juggled more frequently required less interaction time to solve puzzles, we used a linear mixed-effects model on data including only puzzle solves, and excluding puzzle solve failures to avoid ceiling effects. Interaction time was the response variable, juggle rate, species, the interaction between juggle rate and species, age, sex, site and puzzle type were fixed effects, and individual ID nested within group ID were random effects.

Model selection was conducted as described above for the rock juggling observational data. We conducted all statistical analyses in RStudio version 3.6.0.